Re: [Roll-bier-dt] roll-bier-dt

[Toerless Eckert <[e-mail]>]

On Fri, Oct 19, 2018 at 01:20:11PM +0000, Pascal Thubert (pthubert) wrote:
> Sure, but the mystery to me is what goes transported with BIER.

As said, BIER proper can encapsulate anything deemed important enough to
get a code point in the BIER encap. I am saying "proper" because ROLL
may not want to take RFC8296 given the likely competition with 6LowPan.

> Because that is the heart of the problem we are facing in this DT.
> If BIER transports a same IP packet to multiple destinations, then they must all accept the same packet.

More a property of the larger clas of multicasting overall assuming
tojust copy the payload - and not to modify it.

I thought we had killed in last weeks meeting the idea of trying to come
up with a mechanism that manages to use packet replication (BIER) and then
modding received copies to fake different per-receiver IP unicast packets
- because of the problem of making this architecturally work well with 
higher layers like UDP that also have a notion of an IP (pseudo) header.
Too much faking needed. At least thats what i tried to summarize in my slides.

Aka: We use BIER-TE for unicast packets but then the expectation is
that there is just one destination, and the whole purpose of BIER-TE is
not replication but source routing. And we use BIER-TE(/BIER) for IP
multicast packets where there can be 1 or more destinations.

[ Of course there are fun extensions later, e.g.: a unicast packet could
 still be sent to redundant nodes sharing an IP adddress and therefore
 able to receive/process the same packet, but i think thats less likely
 something one would do in a ROLL domain. ]

> Moreover, if the destination is not the router indicated by the bit
> but a neighbor or a set of neighbors to that router, then the inner
> packet must indicate that.
> All in all, that tells me that the transported packet is IP multicast.

Yes, we have two layers. The BIER layer that only reaches nodes with
an assigned BIER bit, and they can communicate potentially any
payload across BIER, IP or other. And then on top of that there
is IP multicast between the BIER edge nodes and IP unicast/multicast
only nodes. The pictures in the slide deck posted intend to show that (that
IP multicast building block picture still TBD).

> What do I miss?

I think primarily me wanting to figure out the encap we use within
the BIER-ROLL domain: RFC8296 vs. 6LowPan+extensions.

Cheers
    Toerless

> Take care,
> 
> Pascal
> 
> > -----Original Message-----
> > From: Toerless Eckert <[e-mail]>
> > Sent: vendredi 19 octobre 2018 15:14
> > To: Pascal Thubert (pthubert) <[e-mail]>
> > Cc: Carsten Bormann <[e-mail]>; [e-mail]
> > Subject: Re: [Roll-bier-dt] roll-bier-dt
> > 
> > On Fri, Oct 19, 2018 at 12:36:06PM +0000, Pascal Thubert (pthubert) wrote:
> > > Hello Toerless:
> > >
> > > Well, there is as Carsten says a multicast support for RPL storing mode, but not
> > for non-storing. This is where Carsten's draft comes in to play.
> > 
> > Sure. As said, i may not have correctly captured what Mcr was explaining,
> > 
> > > In normal BIER, the bits indicate the routers, don't they? So there's an IP
> > packet inside, correct, like in my tunnel mode? But then I expect it is the same
> > packet for all destinations. So how does htat work, is it a mcast packet or do all
> > the destination have a same IP address?
> > 
> > "Proper BIER" (TM) (the stuff done by BIER-WG) has defined its own
> > encapsulation header (RFC8296) that sits on top of MPLS or ethernet and that
> > has its own next-proto field/registry, so it can be followed by IP or anything else
> > that BIER-WG feels deserves a code point in that registry.
> > 
> > Cheers
> >     Toerless
> > 
> > > Pascal
> > >
> > > > -----Original Message-----
> > > > From: Roll-bier-dt <[e-mail]> On Behalf Of
> > > > Toerless Eckert
> > > > Sent: vendredi 19 octobre 2018 14:21
> > > > To: Carsten Bormann <[e-mail]>
> > > > Cc: [e-mail]
> > > > Subject: Re: [Roll-bier-dt] roll-bier-dt
> > > >
> > > > That was me taking notes what Michael said, and maybe i didn't
> > > > transcribe it correctly. Hope Michael is subscribed here to answer.
> > > >
> > > > I think i was more wondering about the suport of compressed IP
> > > > multicast headders in 6LowPan because that is unclear to me and we
> > > > probably need to take one of the DT meetings to detail more what
> > > > extensions we need to do for 6LowPan to support BIER.
> > > >
> > > > Cheers
> > > >     Toerless
> > > >
> > > > On Fri, Oct 19, 2018 at 07:45:50AM +0200, Carsten Bormann wrote:
> > > > > On Oct 19, 2018, at 04:56, Toerless Eckert <[e-mail]> wrote:
> > > > > >
> > > > > > https://raw.githubusercontent.com/toerless/roll-bier/master/note
> > > > > > s/18
> > > > > > 1017-roll-bier-notes.txt
> > > > >
> > > > > What do you mean with ???RPL does not support multicast.????  MOP=3?
> > > > > (That is of course storing mode only, and storing mode doesn???t
> > > > > exist :-), so
> > > > you have a point.
> > > > > That???s why we did ccast, to fill in multicast for non-storing
> > > > > mode.)
> > > > >
> > > > > Grüße, Carsten
> > > >
> > > > --
> > > > ---
> > > > [e-mail]
> > > >
> > > > _______________________________________________
> > > > Roll-bier-dt mailing list
> > > > [e-mail]
> > > > https://www.ietf.org/mailman/listinfo/roll-bier-dt
> > 
> > --
> > ---
> > [e-mail]
> 
> _______________________________________________
> Roll-bier-dt mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/roll-bier-dt

-- 
---
[e-mail]